GPT+ data analysis gives GPT4 access to a jupyter back-end.
I was dealing with a noisy sensor in a factory production line.
I asked GPT to remind me which averaging functions might be usable to smooth out the noise, within the constraint of a PLC with limited memory.
I got GPT to simulate the output from the noisy sensor, apply each averaging function, and supply graphs of inputs and outputs. It was then pretty easy to eyeball which function was most suitable for the task.
I then gave GPT a code-style example and asked it to provide an (IEC 61131-3) Structured Text implementation using that code style. Which it did. This turned out to be pretty close to the final implementation (after careful reading and testing.)
Because it's so cheap to do (time and money wise), I used GPT to generate quite a lot of throw-away code to get to the final result. I probably wouldn't have considered this particular approach if doing it by hand.
This is exactly my experience and what I am trying to get at. I start feeling that maybe I'm missing out, so I go give it a try. My direct experience shows that it spits out gibberish, but then people say things like it's a godsend but don't give any examples besides maybe it generated some HTTP API requests for them.